22OCT

USA Today Co. sues OpenAI, not Microsoft

4 min read

16:20UTC

USA Today Co. wants more than $250m from seven OpenAI entities, and its complaint says OpenAI's output filters shielded only publishers that had already sued.

← Back to USA Today sues OpenAI but not Microsoft Jump to analysis ↓

IndustryAssessed

Key takeaway

USA Today Co. licenses its content to Microsoft and sues OpenAI for more than $250m.

USA Today Co. sued seven OpenAI entities on 8 October in the US District Court for the Southern District of New York, seeking "damages in excess of $250 million"⁠¹. The 79-page complaint, case 1:26-cv-08892, is brought by the parent and 13 subsidiaries that own 19 titles, including the Detroit Free Press and the Arizona Republic. OpenAI has not yet responded.

The complaint says OpenAI's output filters, meant to stop ChatGPT reproducing articles, protected only the publishers that had sued it. It quotes an unnamed Microsoft executive calling that an "accidental cover up"⁠². If a court accepts that account, a lawsuit becomes the way a publisher keeps its copy out of ChatGPT answers.

The filing counts more than 160,000 of USA Today Co.'s entries in WebText, including 83,266 from usatoday.com alone⁠³. WebText is the scraped dataset OpenAI built in 2019 to train GPT-2, a forerunner of the models behind ChatGPT. The complaint also counts over 122 million tokens from its domains in C4⁠⁴. Tokens are the word fragments a language model reads; C4 is a filtered copy of the Common Crawl web archive.

USA Today Co. left Microsoft out of the suit. The company, formerly Gannett, announced an AI licensing agreement with Microsoft on 30 October 2025⁠⁵, yet its own complaint alleges Microsoft supplied OpenAI with a copy of the Bing search index⁠⁶. The New York Times sued both companies, and so did The Seattle Times and Newsday on 4 September⁠⁷. USA Today Co. hired Rothwell, Figg, Ernst & Manbeck, one of the firms representing the Times, whose generative-AI legal bill the Times reports each quarter. In August the group told shareholders that Palantir helps it monetise readers, and its union demanded it drop that deal.

Deep Analysis

In plain English

USA Today Co. owns USA Today and several hundred American local titles. It used to be called Gannett. OpenAI makes ChatGPT. To build tools like ChatGPT, companies feed huge amounts of text from the internet into a computer program so it learns how language works. USA Today Co. says OpenAI copied more than 160,000 of its web pages to do this without paying, and it has gone to a federal court in New York to ask for money. Microsoft works closely with OpenAI and has a separate deal to pay USA Today Co. for its content. Microsoft is not being sued, even though the complaint says Microsoft gave OpenAI data from its Bing search engine.

Deep Analysis

Root Causes

OpenAI documented its early training sets well enough for a plaintiff to count them. WebText, the 2019 set behind GPT-2, and C4 let USA Today Co. measure its own share entry by entry, which is why the complaint leads with 160,000 entries and 122 million tokens rather than with sample outputs.

USA Today Co. is also running three AI relationships at once. It told shareholders in August that Palantir's data platform helps it monetise its audience, and The NewsGuild-CWA demanded it end that partnership four days later. Each vendor sits in a different box: licensee, supplier, defendant.

What could happen next?

Consequence
If discovery confirms that OpenAI's output filters held back only litigants' content, publishers without a licence gain a direct reason to file rather than wait.
Risk
The unnamed Microsoft executive quoted in paragraph 10 could become a witness, pulling Microsoft into a case in which it is not a defendant.

Source Landscape

This story draws on neutral-leaning sources

Primary parallel: The New York Times sued OpenAI and Microsoft together with the same lead counsel, and its parent booked $8.8m of generative-AI litigation costs in the first half of 2026 alone. USA Today Co. is entering the same consolidated docket in the Southern District of New York, with the same firm, against one of the two companies.

Counter-parallel: The Seattle Times and Newsday filed on 4 September against nine OpenAI entities and Microsoft, with seven counts including trademark dilution under Washington and New York law, and named no damages figure. USA Today Co. narrowed the defendants and put a floor of $250m on its claim.

The $250m floor compares with the $4.6m The New York Times Company booked for generative-AI litigation in the second quarter of 2026, up 32.6% on a year earlier. The New York Times Company pays those costs every quarter on the docket USA Today Co. joined on 8 October, before any damages are decided.

USA Today Co. is suing while taking licence income from Microsoft under the agreement announced on 30 October 2025, so its AI revenue and its AI litigation now run through the same accounts.

Consensus view: Rothwell, Figg, Ernst & Manbeck, the firm that drafted the complaint, builds the case on two mechanisms: the removal of copyright management information from copied articles, and an output filter that allegedly suppressed only the content of publishers that had sued (para 10). On that second theory, the filter itself is evidence that OpenAI could identify and hold back a publisher's copy when it chose to.

Counter-view: Klaris Law, acting for The Seattle Times and Newsday five weeks earlier, pleaded the Bing allegation the other way: it named Microsoft as a defendant and alleged, on information and belief, that Microsoft built bespoke web crawlers for OpenAI as well as supplying the Bing index (para 72).

Key tension: Whether USA Today Co. can rely on Microsoft's alleged role in supplying OpenAI's data while keeping Microsoft as a licensee, and what Microsoft's evidence would show if OpenAI pulls it into the case.

First Reported In

Update #13 · USA Today sues OpenAI but not Microsoft

US District Court SDNY (via Reuters)· 22 Oct 2026

Read original →

Different Perspectives

Publicis Groupe

Publicis Groupe credited AI demand for its Connected Media growth in the 13 October third-quarter release, reporting net revenue of €3,808m, up 5.0% organically. It named no AI product, so investors get the AI claim without a product to measure it against.

Educational Broadcasting System

Educational Broadcasting System premiered EBS AI Literature Theatre on 16 October, remaking 11 Korean literary classics with generative AI. Korea's public education broadcaster puts AI-made drama on air under its own name, while the BBC's director-general said he 'got in trouble' for praising an AI Doctor Who episode.

Hakuhodo

Hakuhodo began selling its in-house Image Similarity Checker on 19 October, scoring AI-generated ad creative against similar images on the web and logging who checked what. The Japanese agency is selling the compliance step around generative advertising and leaves the legal judgment with its clients.

Lisa Nandy

Nandy told Sky on 15 October she is minded to issue a public interest intervention notice on its purchase of ITV's Media and Entertainment business, on four grounds including news plurality. She notes Comcast's Sky News undertakings expire in December 2028 and do not cover ITV News or ITN, in which Sky would take a 20% stake.

Australian Broadcasting Corporation

The Australian Broadcasting Corporation asked Parliament's AI committee to extend the News Bargaining Incentive to AI firms and hold them to the copyright, defamation and privacy rules news outlets follow. With SBS telling the same committee that robots.txt is routinely ignored, the public broadcasters are contesting the opt-out Anthropic proposed.

Editorial Perfil

Perfil reported on 14 October that Civil Court No. 57 kept its damages suit against OpenAI and Microsoft out of the federal courts, and it wants a pre-trial inspection of their California servers. Its lawyer says Microsoft Argentina came to mediation without a serious offer, and Perfil seeks payment rather than takedowns.

Written with AI from Lowdown's published reporting. Not individually reviewed by an editor. Editorial standards.